// Bitcasts between 512-bit vector types. Return the original type since
// no instruction is needed for the conversion
let Predicates = [HasAVX512] in {
def : Pat<(v8f64 (bitconvert (v16f32 VR512:$src))), (v8f64 VR512:$src)>;
def : Pat<(v8f64 (bitconvert (v16i32 VR512:$src))), (v8f64 VR512:$src)>;
def : Pat<(v8f64 (bitconvert (v8i64 VR512:$src))), (v8f64 VR512:$src)>;
def : Pat<(v16f32 (bitconvert (v16i32 VR512:$src))), (v16f32 VR512:$src)>;
def : Pat<(v16f32 (bitconvert (v8i64 VR512:$src))), (v16f32 VR512:$src)>;
def : Pat<(v16f32 (bitconvert (v8f64 VR512:$src))), (v16f32 VR512:$src)>;
def : Pat<(v8i64 (bitconvert (v16f32 VR512:$src))), (v8i64 VR512:$src)>;
def : Pat<(v8i64 (bitconvert (v16i32 VR512:$src))), (v8i64 VR512:$src)>;
def : Pat<(v8i64 (bitconvert (v8f64 VR512:$src))), (v8i64 VR512:$src)>;
def : Pat<(v16i32 (bitconvert (v16f32 VR512:$src))), (v16i32 VR512:$src)>;
def : Pat<(v16i32 (bitconvert (v8i64 VR512:$src))), (v16i32 VR512:$src)>;
def : Pat<(v16i32 (bitconvert (v8f64 VR512:$src))), (v16i32 VR512:$src)>;
def : Pat<(v8f64 (bitconvert (v8i64 VR512:$src))), (v8f64 VR512:$src)>;
def : Pat<(v2i64 (bitconvert (v4i32 VR128X:$src))), (v2i64 VR128X:$src)>;
def : Pat<(v2i64 (bitconvert (v8i16 VR128X:$src))), (v2i64 VR128X:$src)>;
def : Pat<(v2i64 (bitconvert (v16i8 VR128X:$src))), (v2i64 VR128X:$src)>;
def : Pat<(v2i64 (bitconvert (v2f64 VR128X:$src))), (v2i64 VR128X:$src)>;
def : Pat<(v2i64 (bitconvert (v4f32 VR128X:$src))), (v2i64 VR128X:$src)>;
def : Pat<(v4i32 (bitconvert (v2i64 VR128X:$src))), (v4i32 VR128X:$src)>;
def : Pat<(v4i32 (bitconvert (v8i16 VR128X:$src))), (v4i32 VR128X:$src)>;
def : Pat<(v4i32 (bitconvert (v16i8 VR128X:$src))), (v4i32 VR128X:$src)>;
def : Pat<(v4i32 (bitconvert (v2f64 VR128X:$src))), (v4i32 VR128X:$src)>;
def : Pat<(v4i32 (bitconvert (v4f32 VR128X:$src))), (v4i32 VR128X:$src)>;
def : Pat<(v8i16 (bitconvert (v2i64 VR128X:$src))), (v8i16 VR128X:$src)>;
def : Pat<(v8i16 (bitconvert (v4i32 VR128X:$src))), (v8i16 VR128X:$src)>;
def : Pat<(v8i16 (bitconvert (v16i8 VR128X:$src))), (v8i16 VR128X:$src)>;
def : Pat<(v8i16 (bitconvert (v2f64 VR128X:$src))), (v8i16 VR128X:$src)>;
def : Pat<(v8i16 (bitconvert (v4f32 VR128X:$src))), (v8i16 VR128X:$src)>;
def : Pat<(v16i8 (bitconvert (v2i64 VR128X:$src))), (v16i8 VR128X:$src)>;
def : Pat<(v16i8 (bitconvert (v4i32 VR128X:$src))), (v16i8 VR128X:$src)>;
def : Pat<(v16i8 (bitconvert (v8i16 VR128X:$src))), (v16i8 VR128X:$src)>;
def : Pat<(v16i8 (bitconvert (v2f64 VR128X:$src))), (v16i8 VR128X:$src)>;
def : Pat<(v16i8 (bitconvert (v4f32 VR128X:$src))), (v16i8 VR128X:$src)>;
def : Pat<(v4f32 (bitconvert (v2i64 VR128X:$src))), (v4f32 VR128X:$src)>;
def : Pat<(v4f32 (bitconvert (v4i32 VR128X:$src))), (v4f32 VR128X:$src)>;
def : Pat<(v4f32 (bitconvert (v8i16 VR128X:$src))), (v4f32 VR128X:$src)>;
def : Pat<(v4f32 (bitconvert (v16i8 VR128X:$src))), (v4f32 VR128X:$src)>;
def : Pat<(v4f32 (bitconvert (v2f64 VR128X:$src))), (v4f32 VR128X:$src)>;
def : Pat<(v2f64 (bitconvert (v2i64 VR128X:$src))), (v2f64 VR128X:$src)>;
def : Pat<(v2f64 (bitconvert (v4i32 VR128X:$src))), (v2f64 VR128X:$src)>;
def : Pat<(v2f64 (bitconvert (v8i16 VR128X:$src))), (v2f64 VR128X:$src)>;
def : Pat<(v2f64 (bitconvert (v16i8 VR128X:$src))), (v2f64 VR128X:$src)>;
def : Pat<(v2f64 (bitconvert (v4f32 VR128X:$src))), (v2f64 VR128X:$src)>;
// Bitcasts between 256-bit vector types. Return the original type since
// no instruction is needed for the conversion
def : Pat<(v4f64 (bitconvert (v8f32 VR256X:$src))), (v4f64 VR256X:$src)>;
def : Pat<(v4f64 (bitconvert (v8i32 VR256X:$src))), (v4f64 VR256X:$src)>;
def : Pat<(v4f64 (bitconvert (v4i64 VR256X:$src))), (v4f64 VR256X:$src)>;
def : Pat<(v4f64 (bitconvert (v16i16 VR256X:$src))), (v4f64 VR256X:$src)>;
def : Pat<(v4f64 (bitconvert (v32i8 VR256X:$src))), (v4f64 VR256X:$src)>;
def : Pat<(v8f32 (bitconvert (v8i32 VR256X:$src))), (v8f32 VR256X:$src)>;
def : Pat<(v8f32 (bitconvert (v4i64 VR256X:$src))), (v8f32 VR256X:$src)>;
def : Pat<(v8f32 (bitconvert (v4f64 VR256X:$src))), (v8f32 VR256X:$src)>;
def : Pat<(v8f32 (bitconvert (v32i8 VR256X:$src))), (v8f32 VR256X:$src)>;
def : Pat<(v8f32 (bitconvert (v16i16 VR256X:$src))), (v8f32 VR256X:$src)>;
def : Pat<(v4i64 (bitconvert (v8f32 VR256X:$src))), (v4i64 VR256X:$src)>;
def : Pat<(v4i64 (bitconvert (v8i32 VR256X:$src))), (v4i64 VR256X:$src)>;
def : Pat<(v4i64 (bitconvert (v4f64 VR256X:$src))), (v4i64 VR256X:$src)>;
def : Pat<(v4i64 (bitconvert (v32i8 VR256X:$src))), (v4i64 VR256X:$src)>;
def : Pat<(v4i64 (bitconvert (v16i16 VR256X:$src))), (v4i64 VR256X:$src)>;
def : Pat<(v32i8 (bitconvert (v4f64 VR256X:$src))), (v32i8 VR256X:$src)>;
def : Pat<(v32i8 (bitconvert (v4i64 VR256X:$src))), (v32i8 VR256X:$src)>;
def : Pat<(v32i8 (bitconvert (v8f32 VR256X:$src))), (v32i8 VR256X:$src)>;
def : Pat<(v32i8 (bitconvert (v8i32 VR256X:$src))), (v32i8 VR256X:$src)>;
def : Pat<(v32i8 (bitconvert (v16i16 VR256X:$src))), (v32i8 VR256X:$src)>;
def : Pat<(v8i32 (bitconvert (v32i8 VR256X:$src))), (v8i32 VR256X:$src)>;
def : Pat<(v8i32 (bitconvert (v16i16 VR256X:$src))), (v8i32 VR256X:$src)>;
def : Pat<(v8i32 (bitconvert (v8f32 VR256X:$src))), (v8i32 VR256X:$src)>;
def : Pat<(v8i32 (bitconvert (v4i64 VR256X:$src))), (v8i32 VR256X:$src)>;
def : Pat<(v8i32 (bitconvert (v4f64 VR256X:$src))), (v8i32 VR256X:$src)>;
def : Pat<(v16i16 (bitconvert (v8f32 VR256X:$src))), (v16i16 VR256X:$src)>;
def : Pat<(v16i16 (bitconvert (v8i32 VR256X:$src))), (v16i16 VR256X:$src)>;
def : Pat<(v16i16 (bitconvert (v4i64 VR256X:$src))), (v16i16 VR256X:$src)>;
def : Pat<(v16i16 (bitconvert (v4f64 VR256X:$src))), (v16i16 VR256X:$src)>;
def : Pat<(v16i16 (bitconvert (v32i8 VR256X:$src))), (v16i16 VR256X:$src)>;
}
//===----------------------------------------------------------------------===//
// AVX-512 - VECTOR INSERT
//
// -- 32x8 form --
let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
def VINSERTF32x4rr : AVX512AIi8<0x18, MRMSrcReg, (outs VR512:$dst),
(ins VR512:$src1, VR128X:$src2, i8imm:$src3),
"vinsertf32x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, EVEX_4V, EVEX_V512;
let mayLoad = 1 in
def VINSERTF32x4rm : AVX512AIi8<0x18, MRMSrcMem, (outs VR512:$dst),
(ins VR512:$src1, f128mem:$src2, i8imm:$src3),
"vinsertf32x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, EVEX_4V, EVEX_V512, EVEX_CD8<32, CD8VT4>;
}
// -- 64x4 fp form --
let neverHasSideEffects = 1, ExeDomain = SSEPackedDouble in {
def VINSERTF64x4rr : AVX512AIi8<0x1a, MRMSrcReg, (outs VR512:$dst),
(ins VR512:$src1, VR256X:$src2, i8imm:$src3),
"vinsertf64x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, EVEX_4V, EVEX_V512, VEX_W;
let mayLoad = 1 in
def VINSERTF64x4rm : AVX512AIi8<0x1a, MRMSrcMem, (outs VR512:$dst),
(ins VR512:$src1, i256mem:$src2, i8imm:$src3),
"vinsertf64x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, EVEX_4V, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT4>;
}
// -- 32x4 integer form --
let neverHasSideEffects = 1 in {
def VINSERTI32x4rr : AVX512AIi8<0x38, MRMSrcReg, (outs VR512:$dst),
(ins VR512:$src1, VR128X:$src2, i8imm:$src3),
"vinserti32x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, EVEX_4V, EVEX_V512;
let mayLoad = 1 in
def VINSERTI32x4rm : AVX512AIi8<0x38, MRMSrcMem, (outs VR512:$dst),
(ins VR512:$src1, i128mem:$src2, i8imm:$src3),
"vinserti32x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, EVEX_4V, EVEX_V512, EVEX_CD8<32, CD8VT4>;
}
let neverHasSideEffects = 1 in {
// -- 64x4 form --
def VINSERTI64x4rr : AVX512AIi8<0x3a, MRMSrcReg, (outs VR512:$dst),
(ins VR512:$src1, VR256X:$src2, i8imm:$src3),
"vinserti64x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, EVEX_4V, EVEX_V512, VEX_W;
let mayLoad = 1 in
def VINSERTI64x4rm : AVX512AIi8<0x3a, MRMSrcMem, (outs VR512:$dst),
(ins VR512:$src1, i256mem:$src2, i8imm:$src3),
"vinserti64x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, EVEX_4V, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT4>;
}
def : Pat<(vinsert128_insert:$ins (v16f32 VR512:$src1), (v4f32 VR128X:$src2),
(iPTR imm)), (VINSERTF32x4rr VR512:$src1, VR128X:$src2,
(INSERT_get_vinsert128_imm VR512:$ins))>;
def : Pat<(vinsert128_insert:$ins (v8f64 VR512:$src1), (v2f64 VR128X:$src2),
(iPTR imm)), (VINSERTF32x4rr VR512:$src1, VR128X:$src2,
(INSERT_get_vinsert128_imm VR512:$ins))>;
def : Pat<(vinsert128_insert:$ins (v8i64 VR512:$src1), (v2i64 VR128X:$src2),
(iPTR imm)), (VINSERTI32x4rr VR512:$src1, VR128X:$src2,
(INSERT_get_vinsert128_imm VR512:$ins))>;
def : Pat<(vinsert128_insert:$ins (v16i32 VR512:$src1), (v4i32 VR128X:$src2),
(iPTR imm)), (VINSERTI32x4rr VR512:$src1, VR128X:$src2,
(INSERT_get_vinsert128_imm VR512:$ins))>;
def : Pat<(vinsert128_insert:$ins (v16f32 VR512:$src1), (loadv4f32 addr:$src2),
(iPTR imm)), (VINSERTF32x4rm VR512:$src1, addr:$src2,
(INSERT_get_vinsert128_imm VR512:$ins))>;
def : Pat<(vinsert128_insert:$ins (v16i32 VR512:$src1),
(bc_v4i32 (loadv2i64 addr:$src2)),
(iPTR imm)), (VINSERTI32x4rm VR512:$src1, addr:$src2,
(INSERT_get_vinsert128_imm VR512:$ins))>;
def : Pat<(vinsert128_insert:$ins (v8f64 VR512:$src1), (loadv2f64 addr:$src2),
(iPTR imm)), (VINSERTF32x4rm VR512:$src1, addr:$src2,
(INSERT_get_vinsert128_imm VR512:$ins))>;
def : Pat<(vinsert128_insert:$ins (v8i64 VR512:$src1), (loadv2i64 addr:$src2),
(iPTR imm)), (VINSERTI32x4rm VR512:$src1, addr:$src2,
(INSERT_get_vinsert128_imm VR512:$ins))>;
def : Pat<(vinsert256_insert:$ins (v16f32 VR512:$src1), (v8f32 VR256X:$src2),
(iPTR imm)), (VINSERTF64x4rr VR512:$src1, VR256X:$src2,
(INSERT_get_vinsert256_imm VR512:$ins))>;
def : Pat<(vinsert256_insert:$ins (v8f64 VR512:$src1), (v4f64 VR256X:$src2),
(iPTR imm)), (VINSERTF64x4rr VR512:$src1, VR256X:$src2,
(INSERT_get_vinsert256_imm VR512:$ins))>;
def : Pat<(vinsert128_insert:$ins (v8i64 VR512:$src1), (v4i64 VR256X:$src2),
(iPTR imm)), (VINSERTI64x4rr VR512:$src1, VR256X:$src2,
(INSERT_get_vinsert256_imm VR512:$ins))>;
def : Pat<(vinsert128_insert:$ins (v16i32 VR512:$src1), (v8i32 VR256X:$src2),
(iPTR imm)), (VINSERTI64x4rr VR512:$src1, VR256X:$src2,
(INSERT_get_vinsert256_imm VR512:$ins))>;
def : Pat<(vinsert256_insert:$ins (v16f32 VR512:$src1), (loadv8f32 addr:$src2),
(iPTR imm)), (VINSERTF64x4rm VR512:$src1, addr:$src2,
(INSERT_get_vinsert256_imm VR512:$ins))>;
def : Pat<(vinsert256_insert:$ins (v8f64 VR512:$src1), (loadv4f64 addr:$src2),
(iPTR imm)), (VINSERTF64x4rm VR512:$src1, addr:$src2,
(INSERT_get_vinsert256_imm VR512:$ins))>;
def : Pat<(vinsert256_insert:$ins (v8i64 VR512:$src1), (loadv4i64 addr:$src2),
(iPTR imm)), (VINSERTI64x4rm VR512:$src1, addr:$src2,
(INSERT_get_vinsert256_imm VR512:$ins))>;
def : Pat<(vinsert256_insert:$ins (v16i32 VR512:$src1),
(bc_v8i32 (loadv4i64 addr:$src2)),
(iPTR imm)), (VINSERTI64x4rm VR512:$src1, addr:$src2,
(INSERT_get_vinsert256_imm VR512:$ins))>;
// vinsertps - insert f32 to XMM
def VINSERTPSzrr : AVX512AIi8<0x21, MRMSrcReg, (outs VR128X:$dst),
(ins VR128X:$src1, VR128X:$src2, u32u8imm:$src3),
"vinsertps{z}\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR128X:$dst, (X86insrtps VR128X:$src1, VR128X:$src2, imm:$src3))]>,
EVEX_4V;
def VINSERTPSzrm: AVX512AIi8<0x21, MRMSrcMem, (outs VR128X:$dst),
(ins VR128X:$src1, f32mem:$src2, u32u8imm:$src3),
"vinsertps{z}\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR128X:$dst, (X86insrtps VR128X:$src1,
(v4f32 (scalar_to_vector (loadf32 addr:$src2))),
imm:$src3))]>, EVEX_4V, EVEX_CD8<32, CD8VT1>;
//===----------------------------------------------------------------------===//
// AVX-512 VECTOR EXTRACT
//---
let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
// -- 32x4 form --
def VEXTRACTF32x4rr : AVX512AIi8<0x19, MRMDestReg, (outs VR128X:$dst),
(ins VR512:$src1, i8imm:$src2),
"vextractf32x4\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, EVEX, EVEX_V512;
def VEXTRACTF32x4mr : AVX512AIi8<0x19, MRMDestMem, (outs),
(ins f128mem:$dst, VR512:$src1, i8imm:$src2),
"vextractf32x4\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, EVEX, EVEX_V512, EVEX_CD8<32, CD8VT4>;
// -- 64x4 form --
def VEXTRACTF64x4rr : AVX512AIi8<0x1b, MRMDestReg, (outs VR256X:$dst),
(ins VR512:$src1, i8imm:$src2),
"vextractf64x4\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, EVEX, EVEX_V512, VEX_W;
let mayStore = 1 in
def VEXTRACTF64x4mr : AVX512AIi8<0x1b, MRMDestMem, (outs),
(ins f256mem:$dst, VR512:$src1, i8imm:$src2),
"vextractf64x4\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, EVEX, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT4>;
}
let neverHasSideEffects = 1 in {
// -- 32x4 form --
def VEXTRACTI32x4rr : AVX512AIi8<0x39, MRMDestReg, (outs VR128X:$dst),
(ins VR512:$src1, i8imm:$src2),
"vextracti32x4\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, EVEX, EVEX_V512;
def VEXTRACTI32x4mr : AVX512AIi8<0x39, MRMDestMem, (outs),
(ins i128mem:$dst, VR512:$src1, i8imm:$src2),
"vextracti32x4\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, EVEX, EVEX_V512, EVEX_CD8<32, CD8VT4>;
// -- 64x4 form --
def VEXTRACTI64x4rr : AVX512AIi8<0x3b, MRMDestReg, (outs VR256X:$dst),
(ins VR512:$src1, i8imm:$src2),
"vextracti64x4\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, EVEX, EVEX_V512, VEX_W;
let mayStore = 1 in
def VEXTRACTI64x4mr : AVX512AIi8<0x3b, MRMDestMem, (outs),
(ins i256mem:$dst, VR512:$src1, i8imm:$src2),
"vextracti64x4\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, EVEX, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT4>;
}
def : Pat<(vextract128_extract:$ext (v16f32 VR512:$src1), (iPTR imm)),
(v4f32 (VEXTRACTF32x4rr VR512:$src1,
(EXTRACT_get_vextract128_imm VR128X:$ext)))>;
def : Pat<(vextract128_extract:$ext VR512:$src1, (iPTR imm)),
(v4i32 (VEXTRACTF32x4rr VR512:$src1,
(EXTRACT_get_vextract128_imm VR128X:$ext)))>;
def : Pat<(vextract128_extract:$ext (v8f64 VR512:$src1), (iPTR imm)),
(v2f64 (VEXTRACTF32x4rr VR512:$src1,
(EXTRACT_get_vextract128_imm VR128X:$ext)))>;
def : Pat<(vextract128_extract:$ext (v8i64 VR512:$src1), (iPTR imm)),
(v2i64 (VEXTRACTI32x4rr VR512:$src1,
(EXTRACT_get_vextract128_imm VR128X:$ext)))>;
def : Pat<(vextract256_extract:$ext (v16f32 VR512:$src1), (iPTR imm)),
(v8f32 (VEXTRACTF64x4rr VR512:$src1,
(EXTRACT_get_vextract256_imm VR256X:$ext)))>;
def : Pat<(vextract256_extract:$ext (v16i32 VR512:$src1), (iPTR imm)),
(v8i32 (VEXTRACTI64x4rr VR512:$src1,
(EXTRACT_get_vextract256_imm VR256X:$ext)))>;
def : Pat<(vextract256_extract:$ext (v8f64 VR512:$src1), (iPTR imm)),
(v4f64 (VEXTRACTF64x4rr VR512:$src1,
(EXTRACT_get_vextract256_imm VR256X:$ext)))>;
def : Pat<(vextract256_extract:$ext (v8i64 VR512:$src1), (iPTR imm)),
(v4i64 (VEXTRACTI64x4rr VR512:$src1,
(EXTRACT_get_vextract256_imm VR256X:$ext)))>;
// A 256-bit subvector extract from the first 512-bit vector position
// is a subregister copy that needs no instruction.
def : Pat<(v8i32 (extract_subvector (v16i32 VR512:$src), (iPTR 0))),
(v8i32 (EXTRACT_SUBREG (v16i32 VR512:$src), sub_ymm))>;
def : Pat<(v8f32 (extract_subvector (v16f32 VR512:$src), (iPTR 0))),
(v8f32 (EXTRACT_SUBREG (v16f32 VR512:$src), sub_ymm))>;
def : Pat<(v4i64 (extract_subvector (v8i64 VR512:$src), (iPTR 0))),
(v4i64 (EXTRACT_SUBREG (v8i64 VR512:$src), sub_ymm))>;
def : Pat<(v4f64 (extract_subvector (v8f64 VR512:$src), (iPTR 0))),
(v4f64 (EXTRACT_SUBREG (v8f64 VR512:$src), sub_ymm))>;
// zmm -> xmm
def : Pat<(v4i32 (extract_subvector (v16i32 VR512:$src), (iPTR 0))),
(v4i32 (EXTRACT_SUBREG (v16i32 VR512:$src), sub_xmm))>;
def : Pat<(v2i64 (extract_subvector (v8i64 VR512:$src), (iPTR 0))),
(v2i64 (EXTRACT_SUBREG (v8i64 VR512:$src), sub_xmm))>;
def : Pat<(v2f64 (extract_subvector (v8f64 VR512:$src), (iPTR 0))),
(v2f64 (EXTRACT_SUBREG (v8f64 VR512:$src), sub_xmm))>;
def : Pat<(v4f32 (extract_subvector (v16f32 VR512:$src), (iPTR 0))),
(v4f32 (EXTRACT_SUBREG (v16f32 VR512:$src), sub_xmm))>;
// A 128-bit subvector insert to the first 512-bit vector position
// is a subregister copy that needs no instruction.
def : Pat<(insert_subvector undef, (v2i64 VR128X:$src), (iPTR 0)),
(INSERT_SUBREG (v8i64 (IMPLICIT_DEF)),
(INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128X:$src, sub_xmm),
sub_ymm)>;
def : Pat<(insert_subvector undef, (v2f64 VR128X:$src), (iPTR 0)),
(INSERT_SUBREG (v8f64 (IMPLICIT_DEF)),
(INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128X:$src, sub_xmm),
sub_ymm)>;
def : Pat<(insert_subvector undef, (v4i32 VR128X:$src), (iPTR 0)),
(INSERT_SUBREG (v16i32 (IMPLICIT_DEF)),
(INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128X:$src, sub_xmm),
sub_ymm)>;
def : Pat<(insert_subvector undef, (v4f32 VR128X:$src), (iPTR 0)),
(INSERT_SUBREG (v16f32 (IMPLICIT_DEF)),
(INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128X:$src, sub_xmm),
sub_ymm)>;
def : Pat<(insert_subvector undef, (v4i64 VR256X:$src), (iPTR 0)),
(INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR256X:$src, sub_ymm)>;
def : Pat<(insert_subvector undef, (v4f64 VR256X:$src), (iPTR 0)),
(INSERT_SUBREG (v8f64 (IMPLICIT_DEF)), VR256X:$src, sub_ymm)>;
def : Pat<(insert_subvector undef, (v8i32 VR256X:$src), (iPTR 0)),
(INSERT_SUBREG (v16i32 (IMPLICIT_DEF)), VR256X:$src, sub_ymm)>;
def : Pat<(insert_subvector undef, (v8f32 VR256X:$src), (iPTR 0)),
(INSERT_SUBREG (v16f32 (IMPLICIT_DEF)), VR256X:$src, sub_ymm)>;
// vextractps - extract 32 bits from XMM
def VEXTRACTPSzrr : AVX512AIi8<0x17, MRMDestReg, (outs GR32:$dst),
(ins VR128X:$src1, u32u8imm:$src2),
"vextractps{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set GR32:$dst, (extractelt (bc_v4i32 (v4f32 VR128X:$src1)), imm:$src2))]>,
EVEX;
def VEXTRACTPSzmr : AVX512AIi8<0x17, MRMDestMem, (outs),
(ins f32mem:$dst, VR128X:$src1, u32u8imm:$src2),
"vextractps{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(store (extractelt (bc_v4i32 (v4f32 VR128X:$src1)), imm:$src2),
addr:$dst)]>, EVEX;
//===---------------------------------------------------------------------===//
// AVX-512 BROADCAST
//---
multiclass avx512_fp_broadcast<bits<8> opc, string OpcodeStr,
RegisterClass DestRC,
RegisterClass SrcRC, X86MemOperand x86memop> {
def rr : AVX5128I<opc, MRMSrcReg, (outs DestRC:$dst), (ins SrcRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[]>, EVEX;
def rm : AVX5128I<opc, MRMSrcMem, (outs DestRC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),[]>, EVEX;
}
let ExeDomain = SSEPackedSingle in {
defm VBROADCASTSSZ : avx512_fp_broadcast<0x18, "vbroadcastss{z}", VR512,
VR128X, f32mem>,
EVEX_V512, EVEX_CD8<32, CD8VT1>;
}
let ExeDomain = SSEPackedDouble in {
defm VBROADCASTSDZ : avx512_fp_broadcast<0x19, "vbroadcastsd{z}", VR512,
VR128X, f64mem>,
EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>;
}
def : Pat<(v16f32 (X86VBroadcast (loadf32 addr:$src))),
(VBROADCASTSSZrm addr:$src)>;
def : Pat<(v8f64 (X86VBroadcast (loadf64 addr:$src))),
(VBROADCASTSDZrm addr:$src)>;
multiclass avx512_int_broadcast_reg<bits<8> opc, string OpcodeStr,
RegisterClass SrcRC, RegisterClass KRC> {
def Zrr : AVX5128I<opc, MRMSrcReg, (outs VR512:$dst), (ins SrcRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[]>, EVEX, EVEX_V512;
def Zkrr : AVX5128I<opc, MRMSrcReg, (outs VR512:$dst),
(ins KRC:$mask, SrcRC:$src),
!strconcat(OpcodeStr,
"\t{$src, $dst {${mask}} {z}|$dst {${mask}} {z}, $src}"),
[]>, EVEX, EVEX_V512, EVEX_KZ;
}
defm VPBROADCASTDr : avx512_int_broadcast_reg<0x7C, "vpbroadcastd", GR32, VK16WM>;
defm VPBROADCASTQr : avx512_int_broadcast_reg<0x7C, "vpbroadcastq", GR64, VK8WM>,
VEX_W;
def : Pat <(v16i32 (X86vzext VK16WM:$mask)),
(VPBROADCASTDrZkrr VK16WM:$mask, (i32 (MOV32ri 0x1)))>;
def : Pat <(v8i64 (X86vzext VK8WM:$mask)),
(VPBROADCASTQrZkrr VK8WM:$mask, (i64 (MOV64ri 0x1)))>;
def : Pat<(v16i32 (X86VBroadcast (i32 GR32:$src))),
(VPBROADCASTDrZrr GR32:$src)>;
def : Pat<(v8i64 (X86VBroadcast (i64 GR64:$src))),
(VPBROADCASTQrZrr GR64:$src)>;
multiclass avx512_int_broadcast_rm<bits<8> opc, string OpcodeStr,
X86MemOperand x86memop, PatFrag ld_frag,
RegisterClass DstRC, ValueType OpVT, ValueType SrcVT,
RegisterClass KRC> {
def rr : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst), (ins VR128X:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set DstRC:$dst,
(OpVT (X86VBroadcast (SrcVT VR128X:$src))))]>, EVEX;
def krr : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst), (ins KRC:$mask,
VR128X:$src),
!strconcat(OpcodeStr,
"\t{$src, ${dst}{${mask}}{z}|${dst}{${mask}}{z}, $src}"),
[(set DstRC:$dst,
(OpVT (X86VBroadcastm KRC:$mask, (SrcVT VR128X:$src))))]>,
EVEX, EVEX_KZ;
def rm : AVX5128I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set DstRC:$dst,
(OpVT (X86VBroadcast (ld_frag addr:$src))))]>, EVEX;
def krm : AVX5128I<opc, MRMSrcMem, (outs DstRC:$dst), (ins KRC:$mask,
x86memop:$src),
!strconcat(OpcodeStr,
"\t{$src, ${dst}{${mask}}{z}|${dst}{${mask}}{z}, $src}"),
[(set DstRC:$dst, (OpVT (X86VBroadcastm KRC:$mask,
(ld_frag addr:$src))))]>, EVEX, EVEX_KZ;
}
defm VPBROADCASTDZ : avx512_int_broadcast_rm<0x58, "vpbroadcastd", i32mem,
loadi32, VR512, v16i32, v4i32, VK16WM>,
EVEX_V512, EVEX_CD8<32, CD8VT1>;
defm VPBROADCASTQZ : avx512_int_broadcast_rm<0x59, "vpbroadcastq", i64mem,
loadi64, VR512, v8i64, v2i64, VK8WM>, EVEX_V512, VEX_W,
EVEX_CD8<64, CD8VT1>;
def : Pat<(v16f32 (X86VBroadcast (v4f32 VR128X:$src))),
(VBROADCASTSSZrr VR128X:$src)>;
def : Pat<(v8f64 (X86VBroadcast (v2f64 VR128X:$src))),
(VBROADCASTSDZrr VR128X:$src)>;
// Provide fallback in case the load node that is used in the patterns above
// is used by additional users, which prevents the pattern selection.
def : Pat<(v16f32 (X86VBroadcast FR32X:$src)),
(VBROADCASTSSZrr (COPY_TO_REGCLASS FR32X:$src, VR128X))>;
def : Pat<(v8f64 (X86VBroadcast FR64X:$src)),
(VBROADCASTSDZrr (COPY_TO_REGCLASS FR64X:$src, VR128X))>;
let Predicates = [HasAVX512] in {
def : Pat<(v8i32 (X86VBroadcastm (v8i1 VK8WM:$mask), (loadi32 addr:$src))),
(EXTRACT_SUBREG
(v16i32 (VPBROADCASTDZkrm (COPY_TO_REGCLASS VK8WM:$mask, VK16WM),
addr:$src)), sub_ymm)>;
}
//===----------------------------------------------------------------------===//
// AVX-512 BROADCAST MASK TO VECTOR REGISTER
//---
multiclass avx512_mask_broadcast<bits<8> opc, string OpcodeStr,
RegisterClass DstRC, RegisterClass KRC,
ValueType OpVT, ValueType SrcVT> {
def rr : AVX512XS8I<opc, MRMDestReg, (outs DstRC:$dst), (ins KRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[]>, EVEX;
}
defm VPBROADCASTMW2D : avx512_mask_broadcast<0x3A, "vpbroadcastmw2d", VR512,
VK16, v16i32, v16i1>, EVEX_V512;
defm VPBROADCASTMB2Q : avx512_mask_broadcast<0x2A, "vpbroadcastmb2q", VR512,
VK8, v8i64, v8i1>, EVEX_V512, VEX_W;
// Mask register copy, including
// - copy between mask registers
// - load/store mask registers
// - copy from GPR to mask register and vice versa
//
multiclass avx512_mask_mov<bits<8> opc_kk, bits<8> opc_km, bits<8> opc_mk,
string OpcodeStr, RegisterClass KRC,
ValueType vt, X86MemOperand x86memop> {
let neverHasSideEffects = 1 in {
def kk : I<opc_kk, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
let mayLoad = 1 in
def km : I<opc_km, MRMSrcMem, (outs KRC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set KRC:$dst, (vt (load addr:$src)))]>;
let mayStore = 1 in
def mk : I<opc_mk, MRMDestMem, (outs), (ins x86memop:$dst, KRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
}
}
multiclass avx512_mask_mov_gpr<bits<8> opc_kr, bits<8> opc_rk,
string OpcodeStr,
RegisterClass KRC, RegisterClass GRC> {
let neverHasSideEffects = 1 in {
def kr : I<opc_kr, MRMSrcReg, (outs KRC:$dst), (ins GRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
def rk : I<opc_rk, MRMSrcReg, (outs GRC:$dst), (ins KRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
}
}
let Predicates = [HasAVX512] in {
defm KMOVW : avx512_mask_mov<0x90, 0x90, 0x91, "kmovw", VK16, v16i1, i16mem>,
VEX, TB;
defm KMOVW : avx512_mask_mov_gpr<0x92, 0x93, "kmovw", VK16, GR32>,
VEX, TB;
}
let Predicates = [HasAVX512] in {
// GR16 from/to 16-bit mask
def : Pat<(v16i1 (bitconvert (i16 GR16:$src))),
(KMOVWkr (SUBREG_TO_REG (i32 0), GR16:$src, sub_16bit))>;
def : Pat<(i16 (bitconvert (v16i1 VK16:$src))),
(EXTRACT_SUBREG (KMOVWrk VK16:$src), sub_16bit)>;
// Store kreg in memory
def : Pat<(store (v16i1 VK16:$src), addr:$dst),
(KMOVWmk addr:$dst, VK16:$src)>;
def : Pat<(store (v8i1 VK8:$src), addr:$dst),
(KMOVWmk addr:$dst, (v16i1 (COPY_TO_REGCLASS VK8:$src, VK16)))>;
}
// With AVX-512 only, 8-bit mask is promoted to 16-bit mask.
let Predicates = [HasAVX512] in {
// GR from/to 8-bit mask without native support
def : Pat<(v8i1 (bitconvert (i8 GR8:$src))),
(COPY_TO_REGCLASS
(KMOVWkr (SUBREG_TO_REG (i32 0), GR8:$src, sub_8bit)),
VK8)>;
def : Pat<(i8 (bitconvert (v8i1 VK8:$src))),
(EXTRACT_SUBREG
(KMOVWrk (COPY_TO_REGCLASS VK8:$src, VK16)),
sub_8bit)>;
}
// Mask unary operation
// - KNOT
multiclass avx512_mask_unop<bits<8> opc, string OpcodeStr,
RegisterClass KRC, SDPatternOperator OpNode> {
let Predicates = [HasAVX512] in
def rr : I<opc, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set KRC:$dst, (OpNode KRC:$src))]>;
}
multiclass avx512_mask_unop_w<bits<8> opc, string OpcodeStr,
SDPatternOperator OpNode> {
defm W : avx512_mask_unop<opc, !strconcat(OpcodeStr, "w"), VK16, OpNode>,
VEX, TB;
}
defm KNOT : avx512_mask_unop_w<0x44, "knot", not>;
def : Pat<(xor VK16:$src1, (v16i1 immAllOnesV)), (KNOTWrr VK16:$src1)>;
def : Pat<(xor VK8:$src1, (v8i1 immAllOnesV)),
(COPY_TO_REGCLASS (KNOTWrr (COPY_TO_REGCLASS VK8:$src1, VK16)), VK8)>;
// With AVX-512, 8-bit mask is promoted to 16-bit mask.
def : Pat<(not VK8:$src),
(COPY_TO_REGCLASS
(KNOTWrr (COPY_TO_REGCLASS VK8:$src, VK16)), VK8)>;
// Mask binary operation
// - KADD, KAND, KANDN, KOR, KXNOR, KXOR
multiclass avx512_mask_binop<bits<8> opc, string OpcodeStr,
RegisterClass KRC, SDPatternOperator OpNode> {
let Predicates = [HasAVX512] in
def rr : I<opc, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src1, KRC:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set KRC:$dst, (OpNode KRC:$src1, KRC:$src2))]>;
}
multiclass avx512_mask_binop_w<bits<8> opc, string OpcodeStr,
SDPatternOperator OpNode> {
defm W : avx512_mask_binop<opc, !strconcat(OpcodeStr, "w"), VK16, OpNode>,
VEX_4V, VEX_L, TB;
}
def andn : PatFrag<(ops node:$i0, node:$i1), (and (not node:$i0), node:$i1)>;
def xnor : PatFrag<(ops node:$i0, node:$i1), (not (xor node:$i0, node:$i1))>;
let isCommutable = 1 in {
defm KADD : avx512_mask_binop_w<0x4a, "kadd", add>;
defm KAND : avx512_mask_binop_w<0x41, "kand", and>;
let isCommutable = 0 in
defm KANDN : avx512_mask_binop_w<0x42, "kandn", andn>;
defm KOR : avx512_mask_binop_w<0x45, "kor", or>;
defm KXNOR : avx512_mask_binop_w<0x46, "kxnor", xnor>;
defm KXOR : avx512_mask_binop_w<0x47, "kxor", xor>;
}
multiclass avx512_mask_binop_int<string IntName, string InstName> {
let Predicates = [HasAVX512] in
def : Pat<(!cast<Intrinsic>("int_x86_"##IntName##"_v16i1")
VK16:$src1, VK16:$src2),
(!cast<Instruction>(InstName##"Wrr") VK16:$src1, VK16:$src2)>;
}
defm : avx512_mask_binop_int<"kadd", "KADD">;
defm : avx512_mask_binop_int<"kand", "KAND">;
defm : avx512_mask_binop_int<"kandn", "KANDN">;
defm : avx512_mask_binop_int<"kor", "KOR">;
defm : avx512_mask_binop_int<"kxnor", "KXNOR">;
defm : avx512_mask_binop_int<"kxor", "KXOR">;
// With AVX-512, 8-bit mask is promoted to 16-bit mask.
multiclass avx512_binop_pat<SDPatternOperator OpNode, Instruction Inst> {
let Predicates = [HasAVX512] in
def : Pat<(OpNode VK8:$src1, VK8:$src2),
(COPY_TO_REGCLASS
(Inst (COPY_TO_REGCLASS VK8:$src1, VK16),
(COPY_TO_REGCLASS VK8:$src2, VK16)), VK8)>;
}
defm : avx512_binop_pat<and, KANDWrr>;
defm : avx512_binop_pat<andn, KANDNWrr>;
defm : avx512_binop_pat<or, KORWrr>;
defm : avx512_binop_pat<xnor, KXNORWrr>;
defm : avx512_binop_pat<xor, KXORWrr>;
// Mask unpacking
multiclass avx512_mask_unpck<bits<8> opc, string OpcodeStr,
RegisterClass KRC1, RegisterClass KRC2> {
let Predicates = [HasAVX512] in
def rr : I<opc, MRMSrcReg, (outs KRC1:$dst), (ins KRC2:$src1, KRC2:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
}
multiclass avx512_mask_unpck_bw<bits<8> opc, string OpcodeStr> {
defm BW : avx512_mask_unpck<opc, !strconcat(OpcodeStr, "bw"), VK16, VK8>,
VEX_4V, VEX_L, OpSize, TB;
}
defm KUNPCK : avx512_mask_unpck_bw<0x4b, "kunpck">;
multiclass avx512_mask_unpck_int<string IntName, string InstName> {
let Predicates = [HasAVX512] in
def : Pat<(!cast<Intrinsic>("int_x86_"##IntName##"_v16i1")
VK8:$src1, VK8:$src2),
(!cast<Instruction>(InstName##"BWrr") VK8:$src1, VK8:$src2)>;
}
defm : avx512_mask_unpck_int<"kunpck", "KUNPCK">;
// Mask bit testing
multiclass avx512_mask_testop<bits<8> opc, string OpcodeStr, RegisterClass KRC,
SDNode OpNode> {
let Predicates = [HasAVX512], Defs = [EFLAGS] in
def rr : I<opc, MRMSrcReg, (outs), (ins KRC:$src1, KRC:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
[(set EFLAGS, (OpNode KRC:$src1, KRC:$src2))]>;
}
multiclass avx512_mask_testop_w<bits<8> opc, string OpcodeStr, SDNode OpNode> {
defm W : avx512_mask_testop<opc, !strconcat(OpcodeStr, "w"), VK16, OpNode>,
VEX, TB;
}
defm KORTEST : avx512_mask_testop_w<0x98, "kortest", X86kortest>;
defm KTEST : avx512_mask_testop_w<0x99, "ktest", X86ktest>;
// Mask shift
multiclass avx512_mask_shiftop<bits<8> opc, string OpcodeStr, RegisterClass KRC,
SDNode OpNode> {
let Predicates = [HasAVX512] in
def ri : Ii8<opc, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src, i8imm:$imm),
!strconcat(OpcodeStr,
"\t{$imm, $src, $dst|$dst, $src, $imm}"),
[(set KRC:$dst, (OpNode KRC:$src, (i8 imm:$imm)))]>;
}
multiclass avx512_mask_shiftop_w<bits<8> opc1, bits<8> opc2, string OpcodeStr,
SDNode OpNode> {
defm W : avx512_mask_shiftop<opc1, !strconcat(OpcodeStr, "w"), VK16, OpNode>,
VEX, OpSize, TA, VEX_W;
}
defm KSHIFTL : avx512_mask_shiftop_w<0x32, 0x33, "kshiftl", shl>;
defm KSHIFTR : avx512_mask_shiftop_w<0x30, 0x31, "kshiftr", srl>;
// Mask setting all 0s or 1s
multiclass avx512_mask_setop<RegisterClass KRC, ValueType VT, PatFrag Val> {
let Predicates = [HasAVX512] in
let isReMaterializable = 1, isAsCheapAsAMove = 1, isPseudo = 1 in
def #NAME# : I<0, Pseudo, (outs KRC:$dst), (ins), "",
[(set KRC:$dst, (VT Val))]>;
}
multiclass avx512_mask_setop_w<PatFrag Val> {
defm B : avx512_mask_setop<VK8, v8i1, Val>;
defm W : avx512_mask_setop<VK16, v16i1, Val>;
}
defm KSET0 : avx512_mask_setop_w<immAllZerosV>;
defm KSET1 : avx512_mask_setop_w<immAllOnesV>;
// With AVX-512 only, 8-bit mask is promoted to 16-bit mask.
let Predicates = [HasAVX512] in {
def : Pat<(v8i1 immAllZerosV), (COPY_TO_REGCLASS (KSET0W), VK8)>;
def : Pat<(v8i1 immAllOnesV), (COPY_TO_REGCLASS (KSET1W), VK8)>;
}
def : Pat<(v8i1 (extract_subvector (v16i1 VK16:$src), (iPTR 0))),
(v8i1 (COPY_TO_REGCLASS VK16:$src, VK8))>;
def : Pat<(v16i1 (insert_subvector undef, (v8i1 VK8:$src), (iPTR 0))),
(v16i1 (COPY_TO_REGCLASS VK8:$src, VK16))>;
def : Pat<(v8i1 (extract_subvector (v16i1 VK16:$src), (iPTR 8))),
(v8i1 (COPY_TO_REGCLASS (KSHIFTRWri VK16:$src, (i8 8)), VK8))>;